public override AnalysisResult2 Analyze <T>(AnalysisSettings analysisSettings, SegmentSettings <T> segmentSettings)
{
var recording = new AudioRecording(segmentSettings.SegmentAudioFile.FullName);
// get indices configuration - extracted in BeforeAnalyze
var acousticIndicesConfig = (RecognizerConfig)analysisSettings.Configuration;
// get a lazily calculated indices function - if you never get the lazy value, the indices will never be calculated
var lazyIndices = this.GetLazyIndices(
recording,
analysisSettings,
segmentSettings,
acousticIndicesConfig.HighResolutionIndices);
// determine imageWidth for output images
int imageWidth = (int)Math.Floor(
recording.Duration.TotalSeconds
/ acousticIndicesConfig.HighResolutionIndices.IndexCalculationDuration);
// execute actual analysis
RecognizerResults results = this.Recognize(
recording,
analysisSettings.Configuration,
segmentSettings.SegmentStartOffset,
lazyIndices,
segmentSettings.SegmentOutputDirectory,
imageWidth);
var analysisResults = new AnalysisResult2(analysisSettings, segmentSettings, recording.Duration);
BaseSonogram sonogram = results.Sonogram;
double[,] hits = results.Hits;
var predictedEvents = results.GetAllEvents();
// double check all the events have the right offset in case it was missed
foreach (var predictedEvent in predictedEvents)
{
predictedEvent.SegmentStartSeconds = segmentSettings.SegmentStartOffset.TotalSeconds;
predictedEvent.SegmentDurationSeconds = recording.Duration.TotalSeconds;
}
analysisResults.Events = predictedEvents.ToArray();
// compress high resolution indices - and save them.
// IF they aren't used, empty values are returned.
if (lazyIndices.IsValueCreated)
{
this.SummarizeHighResolutionIndices(
analysisResults,
lazyIndices.Value,
acousticIndicesConfig.HighResolutionIndices);
}
// write intermediate output if necessary
if (analysisSettings.AnalysisDataSaveBehavior)
{
this.WriteEventsFile(segmentSettings.SegmentEventsFile, analysisResults.Events);
analysisResults.EventsFile = segmentSettings.SegmentEventsFile;
}
if (analysisSettings.AnalysisDataSaveBehavior)
{
this.WriteSummaryIndicesFile(segmentSettings.SegmentSummaryIndicesFile, analysisResults.SummaryIndices);
}
if (analysisSettings.AnalysisDataSaveBehavior)
{
analysisResults.SpectraIndicesFiles =
this.WriteSpectrumIndicesFiles(
segmentSettings.SegmentSpectrumIndicesDirectory,
segmentSettings.Segment.SourceMetadata.Identifier,
analysisResults.SpectralIndices);
}
if (analysisSettings.AnalysisImageSaveBehavior.ShouldSave(analysisResults.Events.Length))
{
string imagePath = segmentSettings.SegmentImageFile.FullName;
const double EventThreshold = 0.1;
var plots = results.Plots ?? new List <Plot>();
//TODO Remove this when we remove AcousticEvent.
var convertedEvents = predictedEvents.Select(ec => ec is AcousticEvent ae ? EventConverters.ConvertAcousticEventToSpectralEvent(ae) : (EventCommon)ec).ToList();
Image image = this.DrawSonogram(sonogram, hits, plots, convertedEvents, EventThreshold);
image.Save(imagePath);
analysisResults.ImageFile = segmentSettings.SegmentImageFile;
// draw a fancy high res index image
// IF indices aren't used, no image is drawn.
if (lazyIndices.IsValueCreated)
{
this.DrawLongDurationSpectrogram(
segmentSettings.SegmentOutputDirectory,
recording.BaseName,
results.ScoreTrack,
lazyIndices.Value,
acousticIndicesConfig.HighResolutionIndices);
}
}
return(analysisResults);
}
public static (List <EventCommon> SpectralEvents, double[] AmplitudeArray, double[] HarmonicIntensityScores) GetComponentsWithHarmonics(
SpectrogramStandard spectrogram,
int minHz,
int maxHz,
int nyquist,
double decibelThreshold,
double dctThreshold,
double minDuration,
double maxDuration,
int minFormantGap,
int maxFormantGap,
TimeSpan segmentStartOffset)
{
// Event threshold - Determines FP / FN trade-off for events.
//double eventThreshold = 0.2;
var sonogramData = spectrogram.Data;
int frameCount = sonogramData.GetLength(0);
int binCount = sonogramData.GetLength(1);
double freqBinWidth = nyquist / (double)binCount;
int minBin = (int)Math.Round(minHz / freqBinWidth);
int maxBin = (int)Math.Round(maxHz / freqBinWidth);
// extract the sub-band
double[,] subMatrix = MatrixTools.Submatrix(spectrogram.Data, 0, minBin, frameCount - 1, maxBin);
//ii: DETECT HARMONICS
// now look for harmonics in search band using the Xcorrelation-DCT method.
var results = CrossCorrelation.DetectHarmonicsInSpectrogramData(subMatrix, decibelThreshold);
// set up score arrays
double[] dBArray = results.Item1;
double[] harmonicIntensityScores = results.Item2; //an array of formant intesnity
int[] maxIndexArray = results.Item3;
for (int r = 0; r < frameCount; r++)
{
if (harmonicIntensityScores[r] < dctThreshold)
{
continue;
}
//ignore locations with incorrect formant gap
int maxId = maxIndexArray[r];
int bandBinCount = maxBin - minBin + 1;
double freqBinGap = 2 * bandBinCount / (double)maxId;
double formantGap = freqBinGap * freqBinWidth;
if (formantGap < minFormantGap || formantGap > maxFormantGap)
{
harmonicIntensityScores[r] = 0.0;
}
}
// smooth the harmonicIntensityScores array to allow for brief gaps.
harmonicIntensityScores = DataTools.filterMovingAverageOdd(harmonicIntensityScores, 3);
//extract the events based on length and threshhold.
// Note: This method does NOT do prior smoothing of the score array.
var harmonicEvents = AcousticEvent.ConvertScoreArray2Events(
harmonicIntensityScores,
minHz,
maxHz,
spectrogram.FramesPerSecond,
spectrogram.FBinWidth,
dctThreshold,
minDuration,
maxDuration,
segmentStartOffset);
var spectralEvents = new List <EventCommon>();
// add in temporary names to the events
// These can be altered later.
foreach (var he in harmonicEvents)
{
var se = EventConverters.ConvertAcousticEventToSpectralEvent(he);
spectralEvents.Add(se);
se.Name = "Harmonics";
//se.ComponentName = "Harmonics";
}
return(spectralEvents, dBArray, harmonicIntensityScores);
}
